Marina Astitha 1, S.T. Rao 2, Jaemo Yang 1, Huiying Luo 1, Inherent uncertainty in the prediction of ozone and particulate matter for NE US 1 Department of Civil & Environmental Engineering, University of Connecticut, Storrs-Mansfield, CT, USA 2 Department of Marine, Earth & Atmospheric Sciences, North Carolina State University, Raleigh, NC, USA 14th Annual CMAS Conference (Oct 5-Oct 7, 2015), Chapel Hill, NC

MOTIVATION The “best-we-can-do” question: what is the lowest bound for errors and variability in predicted atmospheric and air quality variables using the current state-of-the-science numerical models? How do we best use atmospheric modeling systems for assessing the impact of emission reductions? What is the confidence in using source apportionment methods and address compliance with NAAQS and emissions control strategies? This presentation is the pilot study in the view of addressing the above questions and focuses on the impact of inherent uncertainties on air pollutant concentrations and source apportionment.

OUTLINE Model configuration & initial data Variability of atmospheric fields for NE US Variability of air pollutants for continental and NE US Variability of O 3 source apportionment Remarks & future work

Model Configuration RAMS/ICLAMS (Solomos et al ACP; Kushta et al JGR) modules: Online production of desert dust and sea salt emissions (Solomos et al. 2011; Kushta et al. 2014) Rapid Radiative Transfer Model (RRTM) [Mlawer et al., 1997; Iacono et al., 2008] Desert dust and sea-salt radiative effects as a function of sizeand water content (Solomos et al. 2011; Kushta et al. 2014) Explicit treatment of desert dust and sea salt as CCN, GCCN and IN particles (Nenes and Seinfeld, 2003; Fountoukis and Nenes, 2005) Cloud microphysics: Two-moment bulk scheme [Walko et al., 1995; Meyers et al., 1997] with 5 ice condensates species. CAMx v5.40 (Environ, 2011) Air quality simulations: Emissions inventory for 2005 on 0.1x0.1deg resolution (EDGAR-JRC) Gas, aqueous and aerosol phase chemistry (CB-V, ISORROPIA) Dry and wet deposition OSAT

Model Configuration Gridded domains 5km 25km Simulation period: June Vertical levels up to 20km

Selection of the simulation period Hourly O 3 measurements: EPA’s AQS Ozone concentrations 18 June 2006

Meteorological conditions 06/18/ UTC Bermuda High pressure system Low S-SW winds

Initialization data 1.NCEP Global Forecast System (GFS) Analyses (1x1deg) 2.NCEP FNL (Final) Operational Global Analyses data (1x1deg) (uses Global Data Assimilation System (GDAS), which continuously collects observational data from the Global Telecommunications System (GTS) and other sources) 3.European Center for Medium-Range Weather Forecasts (ECMWF) Analyses (1x1deg) Four Dimensional Data Assimilation (FDDA) (analysis nudging) is implemented in all three simulations

Variability in the global analysis fields for NE US FNL, GFS and ECMWF

Variability of wind

Variability of

Variability in the modeled atmospheric fields for NE US

Daytime ventilation coefficient * mixing height) (6am-9am) FNL GFS ECMWF Variability of ventilation coefficient

Nighttime average (09:00pm-05:00am) Variability of wind speed at the steering level FNLGFS ECMWF

Daytime average (08:00am-05:00pm ) FNL GFS ECMWF Variability of cloud cover

FNL GFS ECMWF Variability of accum. precipitation

Model error as a function of time NE domain

Variability of modeled aerosol concentration (natural sources)

Variability of desert dust load FNL GFSECMWF Particle sizes: 8-size bin scheme

Variability of sea salt load FNLGFS ECMWF Particle sizes: Accumulation + coarse mode

Variability of modeled daily maximum O 3

Coefficient of Variation (mean > 0.5ppb) for each simulation day Variability of daily maximum O 3

FNL GFS ECMWF Daily max Ozone concentration (ppb) for June 18, 2006 max=70.5ppb max=74ppbmax=80.5ppb June 18, 2006

Variability of daily maximum O 3 Coefficient of Variation (mean > 0.5ppb) for each simulation day

Source Apportionment Variability Ozone Source Apportionment Technology (OSAT) in CAMx v5.40 (Environ, 2010) Propagation of inherent uncertainty to the geographic source apportionment for maximum surface ozone concentrations Sensitivity Test No1: 13 Geographic source regions (states); 1 emission group; anthropogenic NOx and VOC precursors only

Source Apportionment Variability Ozone precursor contribution to O 3 max by source area ((Ozone from NOx+VOC)/total; percentages) PRELIMINARY RESULTS

Source Apportionment Variability Ozone precursor contribution to O 3 max by source area MA (ppb) PRELIMINARY RESULTS FNL, ~10ppb max GFS, ~7ppb max ECMWF, ~5ppb max Remaining Questions How important is the variability in source apportionment when applied for attainment demonstration and/or emission control strategies? Can we identify a lower bound of errors in the model prediction and thus improve the confidence in model predictions for regulatory assessments?

FEW REMARKS  We have investigated the inherent uncertainty in atmospheric and air quality models by examining the impact of various initial conditions on weather variables and air pollutant concentrations  The most impacted atmospheric fields are precipitation, cloud cover, ventilation coefficient, and sea salt loading  Ozone daily maximum concentration has shown substantial variability (20-30%) that is more pronounced at the coarser resolution simulations  Propagating inherent uncertainty to the source apportionment shows substantial variability of the source attribution to max ozone values  Further research is needed to quantify the confidence that can be placed in modeled concentrations for policy analysis and regulatory assessments

FUTURE WORK  Dynamic evaluation of air quality model predictions from long-term simulations to analyze the features embedded in model outputs and observations  Determine the influence of inherent uncertainty on PMs in the context of model applications for regulatory assessments (emission reduction policies, design values etc)  Explore emission source contributions to downwind exceedances as influenced by changes in initial state  Determine the confidence by using additional source apportionment and/or sensitivity analysis methods

Acknowledgements This work was partially supported by the Northeast Utilities project “Damage Modeling and Forecasting System of the NU Center Bridge- Funding”. Northeast Utilities and Department of Civil and Environmental Engineering, School of Engineering, University of Connecticut (PI:E. Anagnostou, Co-PI: M. Astitha, B. Hartman, M. Rudnicki). Award: $ , 04/15/ /31/2015. Part of this work was also supported by the Center for Environmental Science and Engineering (CESE) at the University of Connecticut ( through the PhD fellowship for Jaemo Yang. Website: and cee-wrf.engr.uconn.eduwww.airmg.uconn.educee-wrf.engr.uconn.edu